Dear Tiago,
The ImCalc approach is really just one way to show overlapping activation maps within a single figure (with activation maps reflecting suprathreshold (sig.) voxels/clusters), it's no statistical test. Finding no overlap for certain voxels/clusters implies that in at least one of the activation maps, there was no sig. activation at the chosen threshold, but it doesn't imply a sig. difference. You might end up with an overlap of 2 voxels or with 2,000, the message is the same = there's an overlap, but you don't have any statistical values refering to the nature of the overlap.
In any case, there's no need to go into detail on how you managed to come up with the figure with regard to the "creative part". Some viewing tools allow to load several binarized activation maps and have options to control whether different overlays are additive or not, others don't, and some do but you can nonetheless obtain "nicer" figures when loading a combined map from e.g. ImCalc. You would just have to state that the different contrasts were thresholded in a certain way, binarized (or shorter, the voxels/clusters surviving your thresholds were taken), and then mapped onto a certain template image, with the colours coding different conditions/type of overlap.
> each contrast have different threshold (estimated through 3dClustSim)
As it's illustrative you can do anything you like. E.g. a fixed, initial voxel threshold -pthr and a certain fixed -athr, resulting in different minimum cluster sizes for different contrast maps is absolutely reasonable. If you think that for whatever reason, it's wise or justified to go with different initial -pthr and/or different -athr for your analysis then it's also fine for the figure.
> How about a reference for the impossibility to of inference on between-subjects effects
This is just a consequence from the Flexible factorial not allowing valid inference on between-subject effects. See a poster by Donald McLaren and colleagues presented at the OHBM 2011 (http://www.martinos.org/~mclaren/ftp/presentations/OHBM2011_v3.pdf ) on that issue, although it doesn't deal with conjunctions specifically. Note that the tutorial by Gläscher and Gitelman on contrast vectors dates back to 2008, which seems to be very popular still, as people have grown up with it and/or as it is more easy to access or to find when searching for tutorials on ANOVA models in SPM (which is why there are that many topics on that issue).
Best
Helmut
|